KR20090018090A - Image data compression device, compression method, program, and image data decompression device, decompression method, and program - Google Patents

Abstract

A quantizer quantizes ''a prediction error'' as a difference between a pixel X to be compressed and a prediction value X'. A variable-length encoder inputs a quantization number obtained from a quantization table of the quantizer and outputs a variable-length code (compressed code). A prediction value candidate calculation module calculates prediction value candidates (1 to k) according to peripheral pixels A, C, B and transmits them to a multiplexer (MPX). A prediction value decision module decides which of the prediction value candidates (1 to k) is to be a predicted value according to an intermediate value correspondence table based on the image format and an intermediate value/image format correspondence table and transmits a control signal to a multiplexer (MPX). The multiplexer (MPX) outputs one of the decided prediction value candidates (1 to k) as a prediction value X'. Thus, by deciding a prediction value in accordance with the image format, it is possible to reduce the prediction error for each image format.

Description

Image data compression device, compression method and program, and image data restoration device, restoration method and program {IMAGE DATA COMPRESSION DEVICE, COMPRESSION METHOD, PROGRAM, AND IMAGE DATA DECOMPRESSION DEVICE, DECOMPRESSION METHOD, AND PROGRAM}

The present invention provides an image data compression apparatus, a compression method, and a program for performing the compression, which can be processed in real time with high image quality, effective for both two kinds of images having different characteristics (natural images such as movies and CG images such as digital maps). And an image data decompression device for decompressing compressed image data, a decompression method, and a program for performing the decompression.

In order to transmit image information including a large amount of information, data compression is required. In particular, in the case of moving images, image data of 30 to 60 frames is transmitted and received within one second, so data compression becomes indispensable.

By the way, as the image information, a natural image represented by a general television image, a movie, or the like, and a CG image (digital image) represented by a map of a car navigation system or the like are known. Generally, a low frequency component is used in a natural image, and a high frequency is used in a digital image. Contains a lot of ingredients. Background Art In recent years, portable terminals including vehicle-mounted terminals and mobile phones have handled both digital images such as maps and natural images such as TVs and movies. In order to efficiently transmit both image data, low frequency There is a demand for an effective data compression method for both components and high frequency components.

As a conventional image data compression method, a first conventional technique is known in which data compression is performed by a Joint Photographic Experts Group (JPEG) and a Moving Picture Experts Group (MPEG) as shown in FIG. The following patent documents 1 and 2 are mentioned as patent documents which belong to this 1st prior art. As shown in FIG. 1, the first conventional technique blocks (normally 8 * 8 pixels) image data 110 (not shown), applies a frequency transform 210 to the block image, and applies DCT coefficients. The quantization 310 is performed for the quantization 310, and the variable length coding 410 for assigning a code according to the frequency of occurrence is transmitted.

In this case, frequency conversion is usually DCT (Discrete Cosine Transfer), and frequency conversion of image data. Since the human eye is sensitive to low frequency components (flat parts of the image), the DCT coefficients for low frequencies are finely quantized and the DCT coefficients for high frequencies are roughly quantized, so that the image quality is not noticeable for natural images. It is possible to compress. However, since the low-frequency components that are easily detected by the human eye are finely quantized, there is no problem in the compression of the natural image, but the deterioration in image quality is noticeable for the high-frequency components such as lines and characters in the map image (CG image). There is a problem that the CG image is not suitable.

As a conventional image data compression method, a second conventional technique using JPEG-LS (Lossless) as shown in Fig. 2 is known. As shown in Fig. 2, when the image data 120 is compressed, the second conventional technology is a MED (Median Edge Detector) predictor (a type of MAP: Median Adaptive Predictor) from the level value of the pixel to be compressed. Is used to perform level value prediction and directly encode the prediction error. In this second prior art, the frequency of occurrence of the prediction error is basically concentrated around 0, so that a short code is assigned to the prediction error near 0 and a long code is assigned to the prediction error having a large value. Although the compression ratio is about 1/3, since image quality is encoded in units of pixels regardless of high and low frequencies, no degradation in image quality occurs even for CG images.

However, in an interlaced image or an image having a low correlation between lines, prediction is difficult and the prediction error tends to increase. As a result, a variable length code with a low appearance frequency (long code length) is assigned, and each process is slow, such as requiring a calculation process at the time of encoding, and thus there is a problem that it is unsuitable for real-time compression processing.

Patent Document 1: Japanese Patent Application Laid-Open No. 2001-061149

Patent Document 2: Japanese Patent Application Laid-Open No. 2000-069479

As for the conversion encoding by DCT such as JPEG and MPEG, the deterioration in image quality of the CG image is noticeable. As for JPEG-LS, the prediction error for the interlaced image and the image long in the horizontal direction becomes large. It is necessary to use a lot of long codes, which deteriorates the compression efficiency.

In view of this, as an image data compression device for a vehicle-mounted terminal that handles both natural and CG images, an interlaced image or a horizontally long image that is compressed to a high-quality CG image and also shown in a natural image such as a movie It is also necessary not to lower the compression efficiency.

SUMMARY OF THE INVENTION In order to solve the above problems, an object of the present invention is to predict a level value of a pixel to be predicted from a level value of neighboring pixels of the pixel to be predicted, and to compress it by quantizing the error of the prediction result; A compression method, a program for performing the compression, an image data decompression device for restoring compressed image data, a decompression method, and a program for performing the restoration are provided.

The image data compression device of the present invention has a predictor, and predicts the pixel level value of the pixel to be compressed by the predictor. In that case, the prediction value is determined based on the pixel level value and the image format around the compression target pixel. By doing so, the prediction error can be reduced for each image format, and the compression efficiency can be improved. In addition, since it is enough to switch the predicted value in accordance with the picture format, the program and circuit configuration at the time of mounting can be simplified.

The image data compression method of the present invention further comprises calculating a plurality of predicted candidates based on the retained pixel values of the pixels A, B, and C, and based on the intermediate value correspondence table and the intermediate value and image format correspondence table. Determining which of the plurality of predicted candidates is a predicted value, and transmitting a control signal. In this way, in each of the prediction value determining module and the prediction value calculating module, it is possible to determine which prediction value to use in the step of performing the parallel processing and calculating the respective prediction value candidates. It becomes possible.

The program for compressing the image data of the present invention is a program for compressing the image data while predicting the pixel level value of the pixel to be compressed, with respect to the pixels around the pixel to be compressed before encoding to a computer. A pixel on the left, A on the C, and B on the left are maintained, and the pixel values of the pixels A, B, and C and the pixel value of the immediately adjacent line are maintained. Calculating a plurality of prediction value candidates based on pixel values of pixels A, B, and C; compressing the plurality of prediction value candidates based on a median value correspondence table and a median value / image format correspondence table to determine and output one prediction value; Calculating a prediction error from the output prediction value and the compression target pixel, inputting the prediction error to a quantization table to obtain a quantization number, and encoding the quantization number. Enter the table to get the compressed code. In this way, since it is possible to determine which predicted value to use in parallel processing at each stage of predictive value determination and predictive value calculation, and at the end of calculating the predictive value candidate, it is possible to perform data compression at high speed by this parallel processing. .

 In addition, since the image data decompression device and the image data decompression method of the present invention restore the image data compressed by the image data compression device described above by reverse operation, similarly to the image data compression device, each image format is predicted. The error can be reduced and the compression efficiency can be increased. In addition, since it is enough to switch the predicted value in accordance with the picture format, the program and circuit configuration at the time of mounting can be simplified.

In addition, since the program for restoring the image data of the present invention performs the operation opposite to the operation by the program for compressing the image data described above, the predicted value determination and the predicted value are similar to those of the program for compressing the image data described above. In each step of the calculation, it is possible to determine which predicted value to use in the step of completing the calculation of each predictive value candidate in parallel, so that data can be restored at high speed by this parallel processing.

BRIEF DESCRIPTION OF THE DRAWINGS It is a figure which shows the 1st prior art which performs data compression by JPEG and MPEG.

FIG. 2 is a diagram showing a second conventional technique of performing data compression by JPEG-LS.

3 is a block diagram showing a configuration of an image data compression device according to an embodiment of the present invention.

4 is a diagram showing an arrangement relationship between a compression target pixel and a peripheral pixel in the image data compression device according to the embodiment of the present invention.

5 is a diagram illustrating a quantization table in a quantizer according to an embodiment of the present invention.

6 is a diagram showing an intermediate value correspondence table in the predicted value determination module of the image data compression device according to the embodiment of the present invention.

It is a figure which shows the intermediate value / image format correspondence table in the predictive value determination module of the image data compression apparatus which concerns on embodiment of this invention.

8 is a diagram showing a correspondence table showing a correspondence relationship between a control signal of the image data compression device according to the embodiment of the present invention and predicted value candidates 1 to k.

9A is a flowchart for explaining the operation of the image data compression (encoding) device according to the embodiment of the present invention.

FIG. 9B is a flowchart for explaining the processing of the "local decoder" in step S008 in FIG. 9A.

Fig. 10 is a diagram showing pixel level values at any point in time of the compression target pixel and the prediction line buffer in the image data compression device according to the embodiment of the present invention.

11 is a diagram showing a quantization result of an interlace predictor in an image data compression device according to an embodiment of the present invention.

12 is a diagram showing a quantization result of a progressive predictor in the image data compression device according to the embodiment of the present invention.

13 is a functional block diagram showing a system configuration of an image data compression device according to an embodiment of the present invention.

14 is a block diagram showing the configuration of an image data decompression device according to an embodiment of the present invention.

It is a figure which shows the inverse quantization table in the inverse quantizer which concerns on embodiment of this invention.

16 is a diagram illustrating an arrangement relationship between a restoration target pixel and a peripheral pixel in the image data restoration device according to the embodiment of the present invention.

It is a figure which shows the median value correspondence table in the predictive value determination module of the image data decompression | restoration apparatus which concerns on embodiment of this invention.

It is a figure which shows the intermediate value / image format correspondence table in the prediction value determination module of the image data decompression | restoration apparatus which concerns on embodiment of this invention.

It is a figure which shows the correspondence table which shows the correspondence of the control signal of the image data decompression | restoration apparatus which concerns on embodiment of this invention, and predicted value candidates 1-k.

20A is a flowchart for explaining the operation of the image data restoration (decoding) apparatus according to the embodiment of the present invention.

20B is a flowchart for explaining the processing of the "local decoder" in step S027 in FIG. 20A.

Fig. 21 is a diagram showing pixel level values at any point in the predictive line buffer in the image data decompression device according to the embodiment of the present invention.

22 is a diagram showing a quantization result of the interlace predictor in the image data decompression device according to the embodiment of the present invention.

23 is a quantization result diagram of a progressive predictor in the image data decompression device according to the embodiment of the present invention.

24 is a functional block diagram showing a system configuration of an image data decompression device according to an embodiment of the present invention.

EMBODIMENT OF THE INVENTION Hereinafter, embodiment of this invention is described, referring drawings.

3 is a block diagram showing the configuration of an image data compression (encoding) device according to an embodiment of the present invention. In Fig. 3, the image data 001 represents a set of pixels to be compressed, and is extracted by an image processing apparatus (not shown) that digitally processes the image. The compression target pixel X 002 represents a pixel to be compressed at this stage. 4 is a diagram showing the arrangement relationship between the compression target pixel X and the peripheral pixels when the image format is an interlaced image or a progressive image. Also, in FIG. 4, the line is shown as a broken line. The quantizer 003 quantizes "prediction error" which is a difference value between the compression target pixel X (002) and the predicted value X '(006).

5 is a quantization table example showing a correspondence between a quantization value and a quantization number with respect to a prediction error (X-X ') according to an embodiment of the present invention. In the quantization table shown in FIG. 5, a prediction error quantization value (simply referred to as a quantization value) and a quantization number are output with the prediction error (X-X ') as an input. It is preferable that the width of the quantization step is smaller when the absolute value of the prediction error is smaller, and wider when the absolute value of the prediction error is larger. In general, when the prediction error is small, the prediction is a flat image that is easy to match. On the contrary, when the prediction error is large, the prediction tends to be near the edge where the prediction is easily missed. When there is an error in the flat portion, it is easy to be detected by the human eye. Therefore, when the prediction error is small, it is necessary to narrow the step width. Although it depends on whether adaptive quantization is to be performed, the minimum step width is preferably 1 to 4. On the other hand, the step width in the case where the prediction error is large is preferably 24 to 32 so that the frequency of occurrence is small, and in order to integrate many prediction errors into one quantization step with all the power to increase the compression efficiency. In addition, when compressing a complicated pattern with many edges, the predetermined compression rate may not be reached. Therefore, if the quantization step width has a plurality of quantization tables having different widths, and the predetermined compression rate is likely to exceed the predetermined compression rate, the overall step width is roughly quantized. You may perform adaptive quantization to select a table. In addition, the correspondence between the quantization value and the quantization number shown in FIG. 5 needs to be the same correspondence on the compression side and the reconstruction side. Therefore, corresponding to Fig. 5, the decompression quantization table (see Fig. 15) indicating the correspondence between the quantization value and the quantization number is provided on the restoration side. This will be described later.

The variable length encoder 004 inputs a quantization number obtained from the quantization table in FIG. 5 and outputs a variable length code. The compressed code buffer 005 temporarily accumulates a variable length code output from the variable length encoder 004, that is, a compressed code.

The predicted value X '006 is a predicted value candidate selected by the multiplexer [MPX] 015 among predicted value candidates 1 to k (k = 6 in the description of the present embodiment) calculated by the predicted value candidate calculating module 011. Is one of. As shown in FIG. 4, the peripheral pixel A 007 is a pixel on the left side of the compression target pixel X 002, that is, a pixel processed before one of the same lines. The predictive line buffer 008 is a buffer for holding about one line of quantization results for prediction, and is composed of, for example, a shift register. As shown in FIG. 4, the peripheral pixel C 009 is a pixel on the compression target pixel X, that is, a pixel of the same column among the lines processed one earlier. The peripheral pixel B 010 is the pixel on the upper left side of the compression target pixel X, that is, the pixel processed before one of the C 009 among the lines previously processed. In the original image, the peripheral pixels C and the peripheral pixels B correspond to pixels on two lines in the interlaced image and pixels on one line in the progressive image in the compression target pixel X.

The predicted value X '006 is one of the predicted value candidates calculated by the predicted value candidate calculating module 011 based on the peripheral pixels A 007, C 009, and B 010, and initially the picture format is progressive. The calculation of the predictive value candidate in the case of an image will be described. As shown in FIG. 4, in the case of a progressive image, correlation of three directions is taken into consideration by using the pixel level value of the pixel A processed before one of the same lines as the pixel level values of the pixels C and B processed before one line. The predicted value candidates of the predicted value X '006 are A, C, and (A + CB).

Next, the calculation of the prediction value candidate when the picture format is an interlaced picture will be described. In the case of an interlaced image as shown in Fig. 4, since even lines and odd lines are transmitted alternately, in order to perform processing on a line-by-line basis without a frame memory or the like, the pixels processed before two lines having low correlation are selected. Will be used. For this reason, the pixel C '(pixel between pixel X and pixel C) between two lines, and B' (pixel A and pixel) without using the pixel level values of pixels C and B processed directly before two lines. Pixel between B) is assumed. In order to process the same as the above-described progressive image, the pixel level value of the pixel C 'is made the interpolation value "(A + C) / 2" of the pixels A and C, and the level value of the pixel B' is the interpolation of the pixels A and B. It calculates using the value "(2A + CB) / 2". In addition to the pixel C 'and the pixel B', the prediction value candidates of the predicted value X '(006) in consideration of correlation in three directions using the pixel level value of the pixel A processed before one of the same lines are A, (A + C) / 2 and (2A + CB) / 2.

In this way, the prediction value candidate calculation module 011 calculates the prediction value candidates 1 to k shown below based on the peripheral pixels A (007), C (009), and B (010) and transmits them to the multiplexer [MPX] 015. do. In other words,

Prediction candidate 1: (A + C) / 2

Prediction candidate 2: A

Prediction candidate 3: (2A + C-B) / 2

Prediction candidate 4: C

Prediction candidate 5: A

Prediction candidate 6: A + C-B

In the above description, the predicted value candidate calculating module 011 is a predicted value in the case of an interlaced image, which is a value (positive integer including 0) in which the values of the coefficients m and n do not all take 0, Three prediction value candidates "A", "(m * A + n * C) / (m + n)", "A +" using coefficients m and n whose values of m + n are i's power of 2 to enable processing It can be set as an intermediate value of (m * A + n * C) / (m + n)-(m * A + n * B) / (m + n) ". In order to obtain such a predicted value, the predicted value candidate calculating module 01 1 uses three predicted value candidates "A" and "(m * A + n * C) / as predicted value candidates 7-9 in addition to the predicted value candidates 1-6. (m + n) "," A + (m * A + n * C) / (m + n)-(m * A + n * B) / (m + n) "is calculated. As a result, the value of k becomes 9. In other words, for an interlaced image in which the coefficients m and n are considered, the value of k becomes 9 since an interlaced image or progressive image in which the above-described coefficients are not considered is added in addition to the image. Furthermore, when the number of picture formats increases, k becomes three times the number of picture formats. Further, as the coefficients m and n, for example, prediction values for interlacing using m = 1 and n = 3, "A", "(A + 3 * C) / 4" and "(4 * A + 3 * C-3 *" Using B) / 4 ", it is possible to calculate the prediction value resistant to the fluctuation of the pixel level value in the vertical direction. That is, an image with strong correlation in the vertical direction is responded by increasing the value of n rather than m. In addition, as a value of the said coefficient in the case of a progressive image, m = 0 and n = 1 are predetermined, and as a result, it is set as the prediction value similar to the conventional MAP predictor.

The image format 012 is provided to an image data compression device viewed from an image processing apparatus (not shown). In general, the image data format is an image having strong pixel correlation in the horizontal direction or an image having strong pixel correlation in the vertical direction. It is indicated as a signal indicating. In other words, the interlaced image is an image having strong pixel correlation in the horizontal direction, and the progressive image is an image having strong pixel correlation in the vertical direction (an image that does not say that the pixel correlation in the horizontal direction is stronger than the correlation in the vertical direction). Is directed. The same applies to the picture format as a signal indicating, for example, the sampling format of the picture data.

The predicted value determining module 013 determines which of the predicted value candidates 1 to k is the predicted value based on the two correspondence tables shown in FIGS. 6 and 7 below, and the multiplexer [MPX] ( 015). FIG. 6 is an intermediate for narrowing to prediction value candidates (1) to (3) based on a result of comparing the magnitudes of pixel level values of peripheral pixels A, B and C by calculating the median value of peripheral pixels A, B and C. FIG. Value mapping table. The correspondence between the predicted value candidates (1) to (3) and the intermediate value shown in FIG. 6 needs to be the same correspondence on the compression side and the reconstruction side. Therefore, corresponding to FIG. 6, the restoration side has an intermediate value correspondence table (see FIG. 17) indicating correspondence between the predicted value candidates (1) to (3) and the intermediate value. This will be described later.

FIG. 7 shows the prediction value candidates (1) to (3) narrowed in FIG. 6, and the image format 012 is an image having strong pixel correlation in the horizontal direction (interlaced image) or an image having strong pixel correlation in the vertical direction (progressive image). Recognition is an intermediate value / image format correspondence table for finally transmitting to the multiplexer [MPX] 015 a control signal 014 indicating which of the predicted value candidates 1 to k are to be used for the predicted value. Output as the control signal 014 is a 3-bit signal of (000 to 101) as shown in the correspondence table of FIG. Further, as described above, the three predicted value candidates " A " and " (m * A + n * C) / (m + n) " for the interlaced image in which the coefficients m and n are considered from the predictive value candidate calculating module 011. , Predictive value candidates (1) to (a) when the calculated output of "A + (m * A + n * C) / (m + n)-(m * A + n * B) / (m + n)" is added. Prediction value candidates " A " and " (m * A + n * C) / (m + n) identified by a control signal (in this case, represented by 4 bits) in an image column having strong horizontal pixel correlation corresponding to (3). "," A + (m * A + n * C) / (m + n)-(m * A + n * B) / (m + n) "is added. In addition, the correspondence between the intermediate value and the image format shown in Fig. 7 needs to be the same correspondence on the compression side and the reconstruction side. Therefore, corresponding to FIG. 7, the restoration side has an intermediate value and image format correspondence table (see FIG. 18) indicating correspondence between the intermediate value and the image format. This will be described later.

The multiplexer [MPX] 015 determines which of the predicted value candidates 1 to k is selected as the predicted value based on the control signal 014. 8 is a correspondence table showing a correspondence relationship between the control signal 014 and the above-described predicted candidates 1-6. That is, the control signal (000) corresponds to the predicted value candidate 1, the control signal (001) corresponds to the predicted value candidate 2, the control signal (010) corresponds to the predicted value candidate 3, and the control signal (011) to the predicted value candidate 4 The control signal 100 corresponds to the predicted value candidate 5, and the control signal 101 corresponds to the predicted value candidate 6. Further, as described above, the three predicted value candidates " A " and " (m * A + n * C) / (m + n) " for the interlaced image in which the coefficients m and n are considered from the predictive value candidate calculating module 011. , The calculation output of "A + (m * A + n * C) / (m + n)-(m * A + n * B) / (m + n)" is added, in the table of FIG. As predicted candidates to use, predicted candidates "A", "(m * A + n * C) / (m + n)", "A + (m * A + n * C) / (m + n)-(m * A + n * B) / (m + n) "is added. In addition, a control signal of a bit width (in this case, a 4-bit signal) capable of identifying three predicted candidates to which the control signal is added is required. The correspondence between the control signal and the predicted candidate candidate shown in FIG. 8 needs to be the same correspondence on the compression side and the reconstruction side. Corresponding to Fig. 8, the restoration side has a correspondence table (see Fig. 19) indicating the correspondence between the control signal and the predicted candidate. This will be described later. In addition, in the present embodiment, the control signal 014 and the control signal 101 of three bits are associated as the control signal 014. However, the control signal of this type is not a control signal if it can correspond to the predicted candidates 1 to 6. It doesn't matter.

In addition, the prediction value determination module 013 may maintain a correspondence table indicating which prediction value to use as a signal indicating a sampling format of the image data, not only whether it is interlace or progressive, but also determine the prediction value based on this.

Although not shown, the predictive value candidate calculating module 011 transmits a control signal indicating the sampling format of the image data from the predictive value determining module 013 to the predictive value candidate calculating module 011 based on the coefficients m and n described above. May be determined.

9A is a flowchart for explaining the operation of the image data compression (encoding) device according to the embodiment of the present invention. FIG. 9B is a flowchart for explaining the processing of the "local decoder" in step S008 in FIG. 9A. 9A and 9B, the step is abbreviated as S.

Before entering the description of the steps in FIG. 9A, it should be noted that this example assumes looping until the processing is completed for all the image data.

S001: The compression target pixel X (002) is obtained from the image data 001.

S002: The prediction value determination module 013 compares the magnitude relationship of the pixel level values in the peripheral pixels A (007), C (009), and B (010). Based on the magnitude relationship and the image format 012, the control signal 014 is transmitted to the multiplexer [MPX] 015 with reference to the correspondence tables shown in Figs.

S003: The prediction value candidate calculation module 011 calculates the prediction value candidates 1 to 6 based on the peripheral pixels A (007), C (009), and B (010). In this case, the predicted value is calculated by the integer operation and the shift operation. In the integer operation, the decimal point of (2A + C-B + 1) / 2 and (A + C + 1) / 2 is discarded. In real-time calculation, the values of (2A + C-B) / 2 and (A + C) / 2 are used as they are without adding one.

S004: The MPX 015 determines the predicted value based on the predicted candidates 1 to 6 and the control signal 014.

S005: The prediction value 006 is subtracted from the pixel level value of the compression target pixel 002 to calculate a prediction error.

S006: Quantize the prediction error with the quantizer 003 to obtain a quantization value and a quantization number.

S007: The variable length encoder 004 generates a variable length code based on the quantization number. This code may be a Golomb code or an arithmetic code.

S008: In order to compress the next pixel, the peripheral pixels A, B, and C and the prediction line buffer 008 are updated (local decoder).

S009: The peripheral pixel C 009 is substituted as the peripheral pixel B 010 of the next compression target pixel.

S010: The peripheral pixel C 009 of the next compression target pixel is obtained from the prediction line buffer 008.

S011: The peripheral pixel A 007 is substituted into the prediction line buffer 008.

S012: The quantized value and the predicted value are added and substituted as the peripheral pixel A 007 of the next compression target pixel.

For example, when the prediction line buffer 008 and the compression target pixel X 002 have values as shown in FIG. 10, the quantization results of the interlaced predictor and the progressive predictor are shown in the tables shown in FIGS. Appears together. When the predictive line buffer 008 and the compression target pixel X 002 take pixel values as shown in Fig. 10, the tables shown in Figs. 11 and 12 are shown in Figs. 9A and 9B described above. It can be derived simply from the operation of the image data compression (encoding) device according to the embodiment of the present invention. In this case, the pixel C 009 at the time of the previous compression becomes the pixel B 010 at the time of this compression, and the previous "prediction value + prediction error quantization value" becomes the pixel A 007 at this time.

Fig. 13 is a functional block diagram showing the system configuration of the image data compression device according to the embodiment of the present invention, in which the above-described contents are functionalized and expressed in blocks. In FIG. 13, the image data compression device according to the embodiment of the present invention inputs image data 102 to be compressed into the compression processing unit 110, and the compression processing unit 110 first inputs the image data 102. Is read out by the reading unit 111 for each line to extract the compression target pixel, and the prediction processing unit 112 shows the intermediate value correspondence table 113 as shown in FIG. With reference to the intermediate value / image format correspondence table 114 as described above, the predicted value candidate calculating unit 115 corresponding to the predicted value candidate calculating module 01 1 of FIG. 3 is used to calculate the predicted value candidate, and the MPX 015 The predictive value determination unit 116 corresponding to the control unit determines a predicted value from among predicted value candidates calculated based on the image format 104, obtains a prediction error from the difference between the determined predicted value and the compression target pixel, and calculates the predicted error from the quantization unit 003. )on Input to the corresponding quantization processing unit 115, the quantization processing unit 117 refers to the quantization table 118 as shown in FIG. 5, obtains a prediction error quantization value and quantization number from the prediction error, inputs a quantization number, The variable length code (compression code) is obtained from the coding processor 119 corresponding to the variable length coder 004 and accumulated as an output of the compression processor 110 to obtain the compressed data 106.

As described above, according to the image data compression apparatus of the present invention, it is possible to reduce the prediction error for each image format by distinguishing and using the prediction values in accordance with the image format, thereby increasing the compression efficiency. In addition, since it is enough to switch the predicted value in accordance with the picture format, the program and circuit configuration at the time of mounting can be simplified.

In the above, the median value of A, B, and C was obtained, and it was determined which prediction value candidate to select based on the median value correspondence table. However, the median value of the prediction value candidate may be obtained after calculating the prediction value candidate.

In the above description, the predicted values for interlacing were (A + C) / 2, A and (2A + CB) / 2, but in the predicted value candidate calculating module 011, coefficient m, With n, "A", "(m * A + n * C) / (m + n)" and "A + (m * A + n * C) / (m + n)-(m * A + n * B) / (m + n) ", and the coefficients m and n are both non-zero values (positive integers including 0), and m + n to enable processing by a shift operation. Is assigned an i-approved value of 2, for example, "A", "(A + 3 * C) / 4" and "(4 * A + 3 * C-3 * B) / 4" as predicted values for interlacing. It is possible to calculate the predicted value resistant to the fluctuation of the pixel level value in the vertical direction by using.

In addition, the same effect is generated even when the prediction value is distinguished by a signal representing the sampling format of the image data, not only an interlaced or progressive image format.

For example, the sampling format of image data stored in a general DVD is usually 4: 2: 2 format. This means that the image size of the Y component (luminance component) is 720x480, and the horizontal image size of the C component (color component) is half of the Y component. For this reason, since the C component is image data having a strong correlation in the vertical direction, the countermeasures m and n can therefore be made by taking n larger than m. In other words, it is possible to prevent the deterioration of the prediction error by increasing the specific gravity with respect to the peripheral pixel with strong longitudinal correlation (by increasing the value of n). For example, for an interlaced image of sampling format 4: 2: 2 and image size 720x480 of image data, in the case of the C component, m = 1 and n = 3 correspond to the image data correlated in the vertical direction. In the case of the image size 720x480, the m component is set to m = 1 and n = 1, so that the correspondence as the image data correlated in the horizontal direction is performed. Regarding the coefficients m and n in the case of image data correlated in the lateral direction, m and n are set to the same value. However, if deterioration of the prediction error is observed, the response is possible by taking the value of m larger than n. As another example, for a progressive image in which the sampling format of the image data is 4: 4: 4 format and the image size is 800x480, m = 0 and n = 1.

In addition, since only the field memory (a buffer for one line) is used and no frame memory is required, the cost can be reduced at the time of mounting.

In addition, in the above description, the predictive value candidate calculation module 011 calculates k predictive value candidates (six in the above description) and multiplexer [MPX] (based on the compressed control signal based on FIGS. 6 and 7 ( 015) selects one prediction value. According to this processing procedure, it is possible to compress the predictive value candidate in FIGS. 6 and 7 and to execute the predictive value calculation at the same timing, thereby improving the processing speed. That is, since each of the predicted value determining module and the predicted value calculating module can be processed in parallel and the calculation of each predicted value candidate is finished, it is possible to determine which predicted value to use. It becomes possible. In the predicted value determining module and the predicted value calculating module, arithmetic processing is performed using a table, so that the program processing can be speeded up.

On the other hand, as an implementation in the case of reducing the construction scale without placing importance on the processing speed, the intermediate value correspondence table of FIG. 6 is excluded from the configuration, and three predicted values are first calculated based on the image format known in advance on the basis of FIG. It is also contemplated that the median of the two predicted values is directly obtained. In addition, as another pattern, a structure for selecting which one to adopt from a plurality of calculated predicted values is determined by determining which predicted candidate formula should be calculated from the intermediate value correspondence table of FIG. 6 and executing only the above-described determined calculation process. A method of cutting down is also considered.

In addition, the image data compression apparatus of the present invention is an image flowing in a stream form in the order from the top to the bottom of the screen in units of frames when the image to be processed is a progressive image, or in units of fields in the case of an interlaced image. The data is to be processed. In addition, the compressed data after compression encoding is packetized and transmitted for each fixed size according to the transmission protocol of the transmission path to which the apparatus is connected.

14 is a block diagram showing the configuration of an image data decompression (decoding) device according to an embodiment of the present invention. In FIG. 14, the compressed code 021 shows the compressed code of the variable length coded image data which is the output of the image data compression (encoding) apparatus according to the embodiment of the present invention shown in FIG. The decoder 022 receives the compressed code of the image data, and outputs a quantization number corresponding to the code. The inverse quantizer 023 takes a quantization number as an input and outputs a quantized value X (024) of the prediction error.

FIG. 15 is an inverse quantization table showing correspondence of prediction error quantization values with respect to the quantization number included in the inverse quantizer 023. FIG. In the inverse quantization table shown in FIG. 15, a prediction error quantization value is output by inputting a quantization number. The correspondence between the prediction error quantization value and the quantization number shown in FIG. 15 needs to be the same correspondence on the reconstruction side and the compression side. Therefore, corresponding to FIG. 15, the compression side is provided with a quantization table (refer FIG. 5) which shows the correspondence of a prediction error quantization value and a quantization number.

FIG. 16 is a diagram showing an arrangement relationship between the restoration target pixel X and the peripheral pixels when the image format 032 is, for example, an interlaced image or a progressive image. Also, in FIG. 15, the line is shown as a broken line.

The prediction error quantization value X (024) is added to the prediction value X '026 to obtain reconstructed image data 025, and may be the peripheral pixel A 027 of the next reconstruction target pixel.

The peripheral pixel A 027 is a pixel on the left side of the restoration target pixel, as shown in FIG. 16. The predictive line buffer 028 is a buffer that holds about one line of the quantization result for prediction, and is configured of, for example, a shift register. As shown in FIG. 16, the peripheral pixel C (029) is a pixel on the restoration target pixel (previous line), and the peripheral pixel B 030 is a pixel on the left upper side (previous line) of the restoration target pixel. .

The predicted value X '026 is one of the predicted value candidates calculated by the predicted value candidate calculating module 031 based on the neighboring pixels A 027, C (029), and B 030, and initially the picture format is progressive. The calculation of the predictive value candidate in the case of an image will be described. As shown in Fig. 16, in the case of a progressive image, correlation in three directions is taken into consideration by using the pixel level value of the pixel A processed before one of the same lines as the pixel level values of the pixels C and B processed before one line. The predicted value candidates of the predicted value X '026 are A, C, and (A + CB).

Next, the calculation of the prediction value candidate when the picture format is an interlaced picture will be described. As shown in Fig. 16, in the case of an interlace image, even lines and odd lines are transmitted alternately, so that two lines of low correlation have been used to perform processing on line units without a frame memory or the like. Done. For this reason, the pixel C 'between the pixel C and the pixel X, and the pixel between the pixel B and the pixel A as a pixel between two lines, without using the pixel level value of the pixels C and B which were processed directly before two lines. The pixel level values of the pixels C 'and B' are interpolated with the interpolation values "(A + C) / 2" of the pixels A and C and the interpolation values of the pixels A and B in order to assume B 'and the same processing as the above-described progressive image. The prediction value candidate of the predicted value X '026 which takes into account correlation of three directions using "(2A + CB) / 2" and using the pixel level value of the pixel A processed one earlier in the same line is A, (A + C) / 2, (2A + CB) / 2.

In this way, the predictive value candidate calculation module 031 uses the predictive value candidates 1 to k described below based on the peripheral pixels A 027, C (029), and B 030 (k = 6 in the description of the present embodiment). ) Is transmitted to the multiplexer [MPX] 035. In other words,

Prediction candidate 1: (A + C) / 2

Prediction candidate 2: A

Prediction candidate 3: (2A + C-B) / 2

Prediction candidate 4: C

Prediction candidate 5: A

Prediction candidate 6: A + C-B

In the above description, the predicted value candidate calculating module 031 is a predicted value in the case of an interlaced image, which is a value (positive integer including 0) in which the values of the coefficients m and n do not all take 0, In order to enable processing, three predicted candidates "A" and "(m * A + n * C) / (m are used as predicted values of an interlaced image by using the coefficients m and n whose m + n values are i's power of 2. + n) "and" A + (m * A + n * C) / (m + n)-(m * A + n * B) / (m + n) ". In order to obtain such a predicted value, the predicted value candidate calculation module 031 is provided with three predicted value candidates " A " and " (m * A + n * C) / " (m + n) "," A + (m * A + n * C) / (m + n)-(m * A + n * B) / (m + n) "is calculated. As a result, the value of k becomes 9. That is, in the case of an interlaced image in which coefficients m and n are considered, the value of k becomes 9 since an interlaced image or progressive image in which the above-described coefficients are not considered is added to the image format other than the image. Furthermore, when the number of picture formats increases, k becomes three times the number of picture formats.

Further, as the coefficients m and n, for example, prediction values for interlacing using m = 1 and n = 3, "A", "(A + 3 * C) / 4" and "(4 * A + 3 * C-3 *" Using B) / 4 ", it is possible to calculate the prediction value resistant to the fluctuation of the pixel level value in the vertical direction. In addition, as a value of the said coefficient in the case of a progressive image, m = 0 and n = 1 are predetermined, and as a result, it is set as the predicted value similar to the conventional MAP predictor.

The image format 032 is provided to an image data decompression device viewed from an image processing apparatus (not shown). In general, the image data format is an image having strong pixel correlation in the horizontal direction or an image having strong pixel correlation in the vertical direction. It is indicated as a signal indicating. In other words, the interlaced image is an image with strong pixel correlation in the horizontal direction, and the progressive image is indicated as an image with strong pixel correlation in the vertical direction. The same applies to the picture format as a signal indicating, for example, the sampling format of the picture data.

The prediction value determination module 033 determines which of the prediction value candidates 1 to k is to be a prediction value based on the two correspondence tables shown in FIGS. 17 and 18 below, and as a control signal 034, a multiplexer [MPX] 035. To send). FIG. 17 shows the intermediate value for narrowing to predicted value candidates (1) to (3) based on the result of comparing the magnitude of pixel level values of peripheral pixels A, B and C by calculating the median value of peripheral pixels A, B and C. FIG. Value mapping table. In response to FIG. 17, the compression side has an intermediate value correspondence table (see FIG. 6) indicating correspondence between the predicted value candidates (1) to (3) and the median value.

FIG. 18 is a distinction between prediction value candidates (1) to (3) narrowed in FIG. 17 and whether the picture format 032 is an interlaced or progressive image, and finally, which of the prediction value candidates 1 to k is used as a prediction value. An intermediate value / image format correspondence table for transmitting the control signal 034 to the multiplexer [MPX] 035. The output as the control signal 034 is a 3-bit signal of (000 to 101) as shown in the correspondence table of FIG. Further, as described above, three prediction value candidates " A " and " (m * A + n * C) / (m + n) " for the interlaced image in which coefficients m and n are considered from the prediction value candidate calculation module 031. , Predictive value candidates (1) to (a) when the calculated output of "A + (m * A + n * C) / (m + n)-(m * A + n * B) / (m + n)" is added. Prediction value candidates " A " and " (m * A + n * C) / (m + n) identified by a control signal (in this case, represented by 4 bits) in an image column having strong horizontal pixel correlation corresponding to (3). "," A + (m * A + n * C) / (m + n)-(m * A + n * B) / (m + n) "is added. In addition, corresponding to FIG. 18, the compression side has an intermediate value and image format correspondence table (see FIG. 7) indicating correspondence between the intermediate value and the image format.

The multiplexer [MPX] 035 determines which of the predicted value candidates 1 to k is selected as the predicted value, based on the control signal 034. 19 is a correspondence table showing a correspondence relationship between the control signal 034 and the predicted candidates 1 to 6 described above. That is, the control signal (000) corresponds to the predicted value candidate 1, the control signal (001) corresponds to the predicted value candidate 2, the control signal (010) corresponds to the predicted value candidate 3, and the control signal (011) to the predicted value candidate 4 The control signal 100 corresponds to the predicted value candidate 5, and the control signal 101 corresponds to the predicted value candidate 6.

Further, as described above, three prediction value candidates " A " and " (m * A + n * C) / (m + n) " for the interlaced image in which coefficients m and n are considered from the prediction value candidate calculation module 031. , The calculation output of "A + (m * A + n * C) / (m + n)-(m * A + n * B) / (m + n)" is added, in the table of FIG. As predicted candidates to use, predicted candidates "A", "(m * A + n * C) / (m + n)", "A + (m * A + n * C) / (m + n)-(m * A + n * B) / (m + n) "is added. In addition, a control signal of a bit width (in this case, a 4-bit signal) capable of identifying three predicted candidates to which the control signal is added is required. Corresponding to FIG. 19, the compression side has a correspondence table (see FIG. 8) indicating correspondence between control signals and predicted candidates. In the present embodiment, the control signal 034 corresponds to the three-bit control signal (000) to the control signal (101). However, as long as the control signal (034) can correspond to the prediction value candidates 1 to 6, the control signal of the present invention may not be a control signal of this type. Does not matter.

In addition, the prediction value determination module 033 may maintain a corresponding table indicating which prediction value to use as a signal indicating a sampling format of the image data, not only whether it is interlace or progressive, but also determine the prediction value based on this.

Although not disclosed, the control signal indicating the sampling format of the image data is transmitted from the predictive value determination module 033 to the predictive value candidate calculating module 031, and based on this, the predictive value candidate calculating module 031 calculates the coefficients m and n described above. May be determined.

20A is a flowchart for explaining the operation of the image data restoration (decoding) apparatus according to the embodiment of the present invention. 20B is a flowchart for explaining the processing of the "local decoder" in step S027 in FIG. 20A. 20A and 20B, the step is abbreviated as S.

Before entering the description of the steps in FIG. 20A, it should be noted that in this example, the premise is to loop until the processing is completed for all the image data.

S021: The code data 021 is input to the decoder 022, and a quantization number is output.

S022: A quantization number is input to the inverse quantizer 023, and a quantization value 024 of the prediction error is output.

S023: The prediction value determination module 033 compares the magnitude relationship of pixel level values in the peripheral pixels A 027, C (029), and B (030). The control signal 034 is transmitted to the multiplexer [MPX] 035 with reference to the correspondence tables shown in FIGS. 17 and 18 on the basis of the magnitude relationship and the image format 032.

S024: The prediction value candidate calculation module 031 calculates the prediction value candidates 1 to 6 based on the peripheral pixels A 027, C (029), and B (030). In this case, the predicted value is calculated by the integer operation and the shift operation. In the integer operation, the decimal point of (2A + C-B + 1) / 2 and (A + C + 1) / 2 is discarded. In real-time calculation, the values of (2A + C-B) / 2 and (A + C) / 2 are used as they are without adding one.

S025: The MPX 035 determines the predicted value based on the predicted value candidates 1 to 6 and the control signal 034.

S026: The quantized value of the prediction error acquired in S022 and the predicted value obtained in S024 are added to calculate a restoration target pixel (restored image data 025).

S027: In order to recover the next pixel, the peripheral pixels A, B, and C and the predictive line buffer 028 are updated (local decoder).

S028: The peripheral pixel C (029) is substituted as the peripheral pixel B 030 of the next restoration target pixel.

S029: Peripheral pixels C (029) of the next reconstruction target pixel are obtained from the predictive line buffer 028.

S030: The peripheral pixel A 027 is substituted into the predictive line buffer 028.

S031: The quantized value and the predicted value are added and substituted as the peripheral pixel A 027 of the next restoration target pixel.

For example, when the predictive line buffer 028 is a value as shown in Fig. 21, the quantization results of the interlace predictor and the progressive predictor are shown in the tables shown in Figs. When the predictive line buffer 028 takes a pixel value as shown in Fig. 21, the table shown in Figs. 22 and 23 shows an image according to the embodiment of the present invention shown in Figs. 20A and 20B. It can be simply derived from the operation of the data recovery (decoding) device. In this case, the pixel C (029) at the time of the previous restoration becomes the pixel B 030 at the time of the restoration, and the previous "prediction value + prediction error quantization value" becomes the pixel A (027) at this time.

24 is a functional block diagram showing the system configuration of the image data decompression device according to the embodiment of the present invention, in which the above-described contents are functionalized and expressed in blocks. In FIG. 24, the image data decompression device according to the embodiment of the present invention inputs the compressed data 202, which is the output of the above-described image data compression device, into the decompression processing unit 210, and the decompression processing unit 210 first inputs the decompression processing unit 210. The image data 202 is input to the decoding processing unit 211 corresponding to the decoder 022 of Fig. 14, and the quantization number is obtained from the decoding processing unit 211. Subsequently, the quantization number is input to the inverse quantization processing unit 212 corresponding to the inverse quantizer 023. The inverse quantization processing unit 212 obtains a prediction error quantization value from the quantization number with reference to the inverse quantization table 213 as shown in FIG. 15. The obtained prediction error quantization value is input to the prediction processing unit 214. The prediction processing unit 214 refers to the intermediate value correspondence table 215 as shown in FIG. 17 and the intermediate value / image format correspondence table 216 as shown in FIG. The predicted value candidate calculator 217 calculates the predicted value candidate, and the predicted value determiner 218 corresponding to the MPX 035 is calculated by the predicted value candidate calculator 217 based on the image format 204. A prediction value is determined from the prediction value candidates. The prediction processing unit 214 outputs the prediction error quantization value calculated by the prediction value candidate calculation unit 217 and the prediction value determined by the prediction value determination unit 218 as output, and restores the output prediction value and the prediction error quantization value. Input to calculation unit 219. The reconstruction value calculator 219 adds the inputted prediction value and the prediction error quantization value to obtain a reconstructed pixel value and accumulates it to obtain reconstructed (image) data 206.

As described above, the image data decompression device of the present invention can reduce the prediction error in each image format by distinguishing and using the predicted values according to the image format, thereby improving the restoration efficiency. In addition, since it is enough to switch the predicted value in accordance with the picture format, the program and circuit configuration at the time of mounting can be simplified.

In the above description, the median of A, B, and C is obtained, and it is determined which prediction value candidate to select based on the median value correspondence table. However, the median value of the prediction value candidate may be obtained after calculating the prediction value candidate.

In the above description, the predicted values for the interlaced image were (A + C) / 2, A and (2A + CB) / 2. However, in the predictive value candidate calculating module, coefficients m and n are calculated from the pixel level values A, B and C. "A", "(m * A + n * C) / (m + n)" and "A + (m * A + n * C) / (m + n)-(m * A + n * B) / (m + n) ", and the coefficients m and n are substituted for any value that does not take all zeros, and" A "," (A + 3 * C) / 4 "and" By using (4 * A + 3 * C-3 * B) / 4 ", it is possible to calculate a prediction value that is resistant to fluctuations in the pixel level value in the vertical direction. In addition, since it is possible to determine which prediction value to use in the step of performing parallel processing in each of the prediction value determination module and the prediction value calculation module and completing the calculation of each prediction value candidate, it is important to perform data restoration at high speed by this parallel processing. It becomes possible.

In addition, the same effect is generated even when the prediction value is distinguished by a signal representing the sampling format of the image data, not only an interlaced or progressive image format.

For example, the sampling format of image data stored in a general DVD is usually 4: 2: 2 format. This means that the image size of the Y component (luminance component) is 720x480, and the horizontal image size of the C component (color component) is half of the Y component. For this reason, since the C component is image data correlated in the vertical direction, it is possible to cope with the coefficients m and n by taking the value of n larger than m. In other words, it is possible to prevent the deterioration of the prediction error by increasing the specific gravity with respect to the peripheral pixel with strong longitudinal correlation (by increasing the value of n). For example, for an interlaced image of sampling format 4: 2: 2 and image size 720x480 of image data, in the case of the C component, m = 1 and n = 3 correspond to the image data correlated in the vertical direction. In the case of the image size 720x480, the m component is set to m = 1 and n = 1, so that the correspondence as the image data correlated in the horizontal direction is performed. Coefficients m and n in the case of image data correlated in the lateral direction can be made by setting m and n to the same value and taking a larger value of m than n if the prediction error is deteriorated. . As another example, in the case of a progressive image in which the sampling format of the image data is 4: 4: 4 format and the image size is 800x480, m = 0 and n = 1.

 In addition, since only the field memory (a buffer for one line) is used and no frame memory is required, the cost in mounting can be reduced. In addition, since only the predicted value is switched, both of the interlace / progressive device can be supported, so that the circuit scale can be simplified.

In addition, in the above description, the predictive value candidate calculating module 031 calculates k predictive value candidates (six in the above description) and multiplexer [MPX] based on the control signal narrowed based on FIGS. 17 and 18. Selects one prediction value. According to this processing procedure, it is possible to narrow the predicted value candidate in FIGS. 17 and 18 and to execute the predicted value calculation at the same timing, thereby improving the processing speed. That is, since each of the predicted value determining module and the predicted value calculating module can be processed in parallel and the calculation of each predicted value candidate is finished, it is possible to determine which predicted value to use. It becomes possible. In the predicted value determining module and the predicted value calculating module, arithmetic processing is performed using a table, so that the program processing can be speeded up.

On the other hand, as an implementation in the case of reducing the construction scale without placing importance on the processing speed, the intermediate value correspondence table of FIG. 17 is excluded from the configuration, and three predicted values are first calculated based on the image format known in advance on the basis of FIG. It is also contemplated that the median of the two predicted values is directly obtained. In addition, as another pattern, a structure for selecting which one to adopt from a plurality of calculated predicted values is determined by determining which predicted candidate expression should be calculated from the intermediate value correspondence table of FIG. A method of cutting down is also considered.

The image data compression device and the image data decompression device of the present invention can be mounted on a desired application device by converting them into a single-chip LSI, for example, to an in-vehicle video data transmission system including car navigation system or to various entertainment (game, animation, etc.) devices. Application is considered. As an example, in the in-vehicle video data transmission system, a display is also provided on the rear side, and it is desired to be able to watch a different image from the front side. Therefore, the image data processed by the video data processing apparatus mounted on the front side is compressed and transferred to the rear side, and the rear side is restored to display the image data. In the case of relaying image data, compression and decompression are repeated at each relay point.

Claims (14)

When compressing the image data, the image data has a predictor for predicting the pixel level value of the pixel to be compressed, and then quantizes the difference between the prediction result of the predictor and the actual pixel level value, and then encodes the image data based on the encoding or the difference value. The compression apparatus, wherein the predictor determines the prediction value based on the pixel level value and the image format around the compression target pixel. 2. The image data compression apparatus as claimed in claim 1, wherein the image format indicates one of an image having strong horizontal pixel correlation and an image having strong vertical pixel correlation. 3. The image data compression device according to claim 2, wherein the image having strong horizontal pixel correlation is an interlaced image, and the image having strong vertical pixel correlation is a progressive image. The image data compression device according to claim 2, wherein any one of the image having strong horizontal pixel correlation and the image having strong vertical pixel correlation is defined by a sampling format. The pixel of claim 2, wherein the left pixel is A, the upper pixel is C, the upper left pixel is B, and the coefficients m and n do not all take 0, and the predicted value is , Three predicted candidates "A", "(m * A + n * C) / (m + n)", "A + (m * A + n * C) / (m + n)-(m * A + and n * B) / (m + n) ". 6. The image data compression according to claim 5, wherein m is 0 and n is 1 when the image data is a progressive image, and m is 1, n is 1 or 3 when the image data is an interlaced image. Device. The pixel to be compressed according to claim 2, wherein the pixel on the left side is A, the pixel on the left side is C, the pixel on the left side is B, the coefficients m and n are not all zeros. An intermediate value of pixels A, B and C around the pixel is obtained. When the intermediate value is A, the predicted value is " A ", and when the intermediate value is C, the predicted value is " (m * A + n * C). / (m + n) ", and when the intermediate value is B, the predicted value is " A + (m * A + n * C) / (m + n)-(m * A + n * B) / (m + n) ". 6. The image data compression device as claimed in claim 5, wherein the coefficients m and n are defined according to the sampling format. 9. The predictor according to any one of claims 1 to 8, wherein the predictor is set to A on the left, C on the left, C on the left, and B on the pixel around the compression target pixel. A prediction value candidate calculation module that calculates a plurality of prediction value candidates based on pixel values of the pixels A, C, and B, and which of the plurality of prediction value candidates is a prediction value based on an intermediate value correspondence table and an intermediate value / image format correspondence table. A prediction value determination module for determining and transmitting as a control signal to the multiplexer, a multiplexer for outputting any one of a plurality of prediction value candidates as a prediction value based on the control signal, and pixel values of the pixels A, B and C in the periphery thereof. And a line buffer for holding pixel values for one line. In the image data compression device having a predictor for predicting pixel level values of a compression target pixel, Before encoding, the pixel on the left side of the compression target pixel is A, the upper pixel is C, the upper left pixel is B, the pixel values of the pixels A, C, and B and just one line in the immediate vicinity. Maintaining the pixel value of Calculating a plurality of candidate candidates based on the pixel values of the surrounding pixels A, C, and B; Determining which of the plurality of predicted candidates is a predicted value based on the intermediate value correspondence table and the intermediate value / image format correspondence table, and transmitting a control signal; Receiving the control signal and outputting any one of a plurality of prediction value candidates as a prediction value; A process of calculating a prediction error from the output prediction value and the compression target pixel, inputting the calculated prediction error to a quantizer to obtain a quantization number, and inputting the quantization number to an encoder to obtain a compressed code. Image data compression method comprising a. A program for compressing image data while predicting a pixel level value of a compression target pixel, the program comprising: Before encoding, the pixel on the left side of the compression target pixel is A, the upper pixel is C, the upper left pixel is B, the pixel values of the pixels A, C, and B and just one line in the immediate vicinity. Maintaining the pixel value of, Calculating a plurality of prediction value candidates based on the retained pixel values of the pixels A, C, and B, Compressing the plurality of candidate candidates based on the intermediate value correspondence table and the intermediate value / image format correspondence table to determine and output one prediction value; Calculating a prediction error from the output prediction value and the compression target pixel, inputting the prediction error to a quantization table to obtain a quantization number, and inputting the quantization number to a coding table to obtain a compressed code Program to run. An image data decompression device for decompressing image data compressed by an image data compression device having a predictor for determining a prediction value based on a pixel level value and an image format around a compression target pixel, comprising: a pixel level value and an image around a compression target pixel A predictor that determines a prediction value based on a format, and adds the predicted value determined by the predictor and the prediction error quantization value obtained by the decoder and the dequantizer to obtain reconstructed pixel data and obtains the next neighboring pixel value. An image data restoration apparatus, characterized in that. An image data decompression device for decompressing image data compressed by an image data compression device having a predictor that determines a prediction value based on a pixel level value and an image format around a compression target pixel, Before restoring the compressed code, the pixel on the left side of the pixel to be restored is A, the upper pixel is C, the upper left pixel is B, the pixel values of the pixels A, C, and B and the Maintaining the pixel value for the vicinity of one line; Calculating a plurality of candidate candidates based on the pixel values of the surrounding pixels A, C, and B; Determining which of the plurality of predicted candidates is a predicted value based on the intermediate value correspondence table and the intermediate value / image format correspondence table, and transmitting a control signal; Receiving the control signal and outputting any one of a plurality of prediction value candidates as a prediction value; Inputting the compressed code data to the inverse encoder to obtain a quantization number, inputting the quantization number to the inverse quantizer to obtain a prediction error quantization value, and restoring a pixel value from the obtained prediction error quantization value and the prediction value Image data restoration method comprising a. A program for decompressing image data compressed by an image data compression device having a predictor that determines a prediction value based on a pixel level value and an image format around a pixel to be compressed, the computer comprising: Before restoring the compressed code, the pixel on the left side of the pixel to be restored is A, the upper pixel is C, the upper left pixel is B, the pixel values of the pixels A, C, and B and the Maintaining pixel values for the vicinity of one line; Calculating a plurality of prediction value candidates based on the retained pixel values of the pixels A, C, and B, Compressing the plurality of candidate candidates based on the intermediate value correspondence table and the intermediate value / image format correspondence table to determine and output one prediction value; Inputting the compressed code data to an inverse encoder to obtain a quantization number, inputting the quantization number to the inverse quantizer to obtain a prediction error quantization value, and restoring a pixel value from the obtained prediction error quantization value and the prediction value Program to run.

Images